| Lesson 2 | Understanding backup sets |
| Objective | Review the structure and purpose of an RMAN backup set. |
Recovery Manager (RMAN) can create two principal physical backup formats: backup sets and image copies. Both formats can protect an Oracle database, but they organize backup data differently and support different storage and recovery strategies.
An RMAN image copy has the same logical structure as the data file, archived redo log, or control file from which it was created. A backup set, by contrast, is an RMAN-specific logical container. RMAN stores its contents in one or more physical backup pieces and must interpret those pieces during a restore. This lesson reviews the backup-set model and explains why RMAN's integration with Oracle Database is important.
The original version of this lesson treated an image copy and a file copied from an operating-system prompt as if they were the same operation. The resulting files can be physically similar, but their creation and management are different. A current recovery plan should distinguish among three concepts:
BACKUP AS COPY and records its metadata in the RMAN repository.When RMAN creates an image copy, it coordinates the operation with the database and records the source file, checkpoint, system change number (SCN), completion time, status, and physical location. For example:
BACKUP AS COPY DATAFILE 7;
RMAN can then list, catalog, crosscheck, validate, switch to, restore, or recover the copy as appropriate. An RMAN image copy can be created while the database is open without placing its data files in user-managed backup mode.
An operating-system or storage-level copy is a user-managed backup. RMAN does not know it exists until a compatible copy is registered with
CATALOG. Cataloging records metadata; it does not repair or validate the copy.
RMAN communicates with the Oracle database while it reads database blocks. This block-aware integration allows RMAN to create a consistent online
backup without surrounding the operation with user-managed BEGIN BACKUP and END BACKUP commands. RMAN also records the information
required to locate and evaluate the backup during restore and recovery.
A simple OS copy command does not provide the following RMAN capabilities by itself:
RMAN does not eliminate the need for a recovery design. Administrators must still protect the repository, archived redo logs, control file, SPFILE, and encryption keys, keep backup pieces accessible, and test the recovery procedure.
Oracle AI Database 26ai still supports user-managed online backups. The technique is relevant when an external storage snapshot or file-copy utility must capture an online data file without using RMAN's block-aware read process. Depending on the scope and storage procedure, the documented workflow may require the database, PDB, or tablespace to be placed in backup mode before the external copy begins.
A database-level user-managed workflow can begin backup mode with:
ALTER DATABASE BEGIN BACKUP;
The administrator or storage system then creates the required copy or snapshot. After every required operation has completed successfully, backup mode must be ended promptly:
ALTER DATABASE END BACKUP;
Backup mode does not make the database read-only and does not stop Oracle from writing to its data files. An external copy can read part of an Oracle block before a database write and another part after that write, producing a split or fractured block in the copy. While backup mode is active, Oracle generates the additional redo information required to reconstruct affected blocks during recovery. Leaving files in backup mode longer than necessary can therefore increase redo generation and operational risk.
A consistent offline user-managed copy can instead be made after a clean database shutdown or, where supported, after taking the relevant tablespace offline. The exact method depends on whether the operation protects a container database (CDB), a pluggable database (PDB), or selected tablespaces and data files. Any production implementation should follow current Oracle 26ai documentation and the storage vendor's validated snapshot procedure.
This lesson no longer provides generic Linux cp or Windows copy commands. The command that copies a file is the simplest part of
the process. Consistency, metadata, archived redo, encryption keys, destination durability, and a tested recovery procedure determine whether the copy is
a usable backup.
A backup set is the smallest logical unit of an RMAN backup. It contains the backup data selected from one or more input files. RMAN records each set in the target control-file repository and, when the environment uses one, in the recovery catalog.
A backup piece is the physical output written by an RMAN channel. A backup set contains one or more backup pieces. On disk, a backup piece appears as a file with an RMAN-specific format. Through a System Backup to Tape (SBT) channel, the media-management or supported cloud library stores the piece as an object that it manages through the SBT interface.
| Object | Role | Management |
|---|---|---|
| Input file | Supplies data-file blocks, control-file data, SPFILE data, or archived redo | Oracle Database and RMAN |
| Backup set | Logical RMAN unit that groups selected backup data | RMAN repository |
| Backup piece | Physical file or media object containing backup-set data | RMAN channel and destination |
| Repository record | Connects the backup to files, checkpoints, pieces, status, and location | Target control file and optional recovery catalog |
The creation sequence is straightforward:
During restore, RMAN queries this metadata to select usable backups, locates the required pieces, and reconstructs the requested database files. A backup piece is not a byte-for-byte data-file copy that an administrator can place directly in the database directory. RMAN must read and interpret it.
RMAN separates data-file content from archived redo-log content. A data-file backup set can contain full or incremental data-file backups and can include a control-file or server-parameter-file backup when allowed by the command. Archived redo logs are stored in archived-log backup sets rather than mixed into the same backup set as data files.
This distinction reflects their recovery roles. A restored data file provides a starting point. Archived and online redo then apply the changes required to recover that file to the desired SCN, time, or log sequence. RMAN uses repository records and data-file headers to determine which backup and redo are needed for the requested recovery.
A backup set can span multiple pieces, and all required pieces must remain available. Backup duplexing, replicated media, and off-site copies can reduce the risk of losing a required piece.
RMAN understands Oracle data-file blocks. When applicable, a data-file backup set can omit blocks that do not need to be backed up, including eligible blocks that have never been used. Oracle documentation describes this behavior as unused-block compression or precompression block processing. It reduces the amount of input written to the backup piece without running a general binary compression algorithm over every byte.
Optional binary compression is a separate backup-set feature. It can be requested for a single backup with:
BACKUP AS COMPRESSED BACKUPSET DATABASE;
Binary compression encodes the selected backup data before RMAN writes it. Its effectiveness depends on the source data, encryption state, selected algorithm, and available CPU capacity. Data that is already compressed or encrypted may provide less additional reduction. Administrators should compare backup duration, CPU use, backup size, and restore speed with representative workloads.
The correct conceptual distinction is:
An RMAN channel represents a stream of work between RMAN and a storage device. A disk channel writes pieces to disk, while an SBT channel passes pieces to a media-management or supported cloud library. Multiple channels can create different backup sets or sections concurrently when the source system and destination can sustain the additional workload.
RMAN can multiplex blocks from multiple input files into a backup piece. Multiplexing helps keep an output device busy when one input file cannot supply
data quickly enough. Multisection backup addresses a different problem by dividing a large data file into independently processed sections. Later lessons
explain how channel parallelism, multiplexing, FILESPERSET, MAXOPENFILES, and section size affect backup and restore throughput.
More channels do not guarantee a proportionally faster backup. Tune parallelism with measured backup and restore performance across the complete storage path.
Oracle AI Database 26ai uses the multitenant architecture. RMAN can create backups for an entire CDB, selected PDBs, tablespaces, or data files, subject to the connection context and command. Backup-set metadata retains the identity and recovery information needed to associate protected files with their containers.
A complete strategy also protects the control file, SPFILE, archived redo logs, RMAN repository, and any encryption keystore. Intact backup pieces cannot be recovered when a required encryption key is unavailable.
Administrators should inspect backup sets through RMAN rather than attempting to open backup pieces with operating-system tools. The following commands show a compact inventory and more detailed information about database backups:
LIST BACKUP SUMMARY;
LIST BACKUP OF DATABASE;
The detailed output identifies logical backup sets and their associated pieces. A repository record, however, does not prove that every piece is still readable or accessible. Files can be deleted outside RMAN, SBT credentials can expire, cloud policies can change, and encryption keys can become unavailable.
Use crosschecks and validation during routine maintenance, then perform periodic restore and recovery tests. A test exercises repository access, media libraries, credentials, encryption keys, network routes, capacity, and the operator procedure.
BEGIN BACKUP and END BACKUP commands.Backup sets remain a core RMAN format in Oracle AI Database 26ai. Their value comes from RMAN's ability to understand Oracle blocks, coordinate online backup activity, store recovery metadata, and manage the pieces required to restore protected files.
In the next lesson, you will examine the BACKUP command syntax used to create backup sets.